Search

Maritime transport forms the backbone of international logistics, as it allows for the transfer of bulk and long-haul products. The sophisticated planning required for this form of transportation frequently involves challenges such as unpredictable weather, diverse types of cargo kinds, and changes in port conditions, all of which can raise operational expenses. As a result, the accurate projection of a ship’s total time spent in port, and the anticipation of potential delays, have become critical for effective port activity planning and management. In this work, we aim to develop a port management system based on enhanced prediction and classification algorithms that are capable of precisely forecasting the lengths of ship stays and delays. On both the training and testing datasets, the XGBoost model was found to consistently outperform the alternative approaches in terms of RMSE, MAE, and R2 values for both the turnaround time and waiting period models. When used in the turnaround time model, the XGBoost model had the lowest RMSE of 1.29 during training and 0.5019 during testing, and also achieved the lowest MAE of 0.802 for training and 0.391 for testing. It also had the highest R2 values of 0.9788 during training and 0.9933 during testing. Similarly, in the waiting period model, the XGBoost model outperformed the random forest and decision tree models, with the lowest RMSE, MAE, and greatest R2 values in both the training and testing phases.

With the increasing demand for efficient maritime logistic management, industries are striving to develop automation software. However, collecting data for analytics from diverse sources like shipping routes, weather conditions, historical incidents, and cargo specifications has become a challenging task in the distribution environment. This challenge gives rise to the possibility of faulty products and traditional testing techniques fall short of achieving optimal performance. To address this issue, we propose a novel decentralised software system based on Transfer Learning and blockchain technology named as BETL (Blockchain -Enabled Transfer Learning). Our proposed system aims to automatically detect and prevent vulnerabilities in maritime operational data by harnessing the power of transfer learning and smart contract-driven blockchain. The vulnerability detection process is automated and does not rely on manually written rules. We introduce a non-vulnerability score range map for the effective classification of operational factors. Additionally, to ensure efficient storage over the blockchain, we integrate an InterPlanetary File System (IPFS). To demonstrate the effectiveness of transfer learning and blockchain integration for secure logistic management, we conduct a testbed-based experiment. The results show that this approach can achieve high precision (98.00%), detection rate (98.98%), accuracy (97.90%), and F-score (98.98), which highlights its benefits in enhancing the safety and reliability of maritime logistics processes. Additionally, the computational time of BETL (the proposed approach) was improved by 18.9% compared to standard transfer learning.

This study incorporates two unique machine learning algorithms, Huber regression and Light Gradient Boosting Machines (LGBM), for estimating ship consumption of fuel. These methods are employed to create forecasting models for ship fuel consumption during journeys, which is especially useful when interacting with non-linear data. The study then analyzes and evaluates the prediction accuracy of these two approaches compared to a baseline model generated using linear regression. The results of the investigation show that both methods establish extremely accurate predictions while handling non-linear data quickly. However, the Huber-based model outperforms the LGBM in terms of prediction accuracy, with an R-squared value of 0.979 versus 0.917 for the LGBM. In addition, the Huber-based model has a diminished prediction error, with an RMSE of 2.278, compared to the LGBM model's RMSE of 4.55. The graphical methods of the violin plot and Taylor’s diagram further established the superiority of Huber ML. These findings imply that Huber regression could be a suitable option for estimating in-route ship fuel usage in real time. As a consequence, this study emphasises the potential benefits of machine learning for accurately predicting ship fuel consumption, providing encouraging possibilities to optimise fuel usage while lowering greenhouse gas emissions. [ABSTRACT FROM AUTHOR]